Music conferences tend to follow a similar format: showcases of up-and-coming artists, panels about the business, and behind-the-scenes deals. But between the events and around the corridors of this year’s Bigsound in Brisbane, Australia, there was one topic dominating conversation: how AI was threatening the industry.
Musicians and composers are fascinated by – and terrified of – artificial intelligence, which has the potential to both help artists create, and steal their work. AI is already starting to weave into the everyday soundtracks of our lives, from a new track by the Beatles to Spotify’s “AI DJ”. It’s a new reality that scholars such as Oliver Bown from the University of New South Wales call “the robot soundscape”.
The tools are already here, and they’re only getting more powerful. AI pioneer OpenAI has MuseNet, a “deep neural network” that uses the same technology as Chat GPT to “generate 4-minute musical compositions with ten different instruments”; while Google has Magenta Studio, an Ableton Live plugin targeting home producers, using “cutting-edge machine learning techniques for music generation”.
Meanwhile, so-called “stem separation” tools make it possible to split digital audio files into their “stems” or harmonic components, such as vocals, bass and drums – a holy grail previously only available to producers with access to master recordings. (Stem separation was the tool that enabled Paul McCartney’s producers to extract John Lennon’s vocals and piano from an old tape of background instruments.)
These tools mean nearly anyone will be able to produce professional-sounding original compositions – although the use of the word “original” is a topic of fierce debate, which made headlines recently when Heart on My Sleeve, a deep-fake song purporting to mimic Drake and the Weeknd, was submitted to the Grammys.
At a panel discussion on music AI at BigSound, the majority of attenders were already using AI tools in their production activities, at least according to a show of hands. Panellist Simon Franglen, a composer, has worked with The Weeknd – and while conceding that Heart on my Sleeve was not yet a convincing dupe, the technology is rapidly improving, he said. “What happens this year is not the issue; what happens in three or five years is.”
Franglen predicts record labels and music publishers will look to “remonetise” dead artists, for instance by creating “new” songs by Frank Sinatra. “The deep fake side of this – you’re not going to know the difference.”
The employment impacts could be significant, including in composition and production. “We are deluding ourselves if we think it isn’t going to happen,” Franglen said. “AI will kill probably all reality TV music within three years … and the cop show where someone is chasing someone down an alley, all of that [music] will be generated.”
Regulation hasn’t caught up with the technology, causing deep concern within the cultural sector. Australia’s representative union, the Media Arts and Entertainment Alliance, has warned that “it is conceivable that many of our members could be replaced by various versions of generative AI”.
skip past newsletter promotion
Sign up to Saved for Later
Catch up on the fun stuff with Guardian Australia’s culture and lifestyle rundown of pop culture, trends and tips
after newsletter promotion
On the other side, the Guardian understands that tech companies and the content industries are both lobbying the federal government for favourable copyright reforms and exemptions that will give them freer access to AIs. At stake is the huge amount of copyrighted compositions tech firms have trawled to train their models. Attorney general Mark Dreyfus’s department is convening a copyright roundtable, and high-profile lawsuits have already begun. “It’s never been legal,” SoundCloud’s Hazel Savage said at BigSound, noting that she had signed specific deals to license music to train large music models. “The industry is still playing catch-up on what that licensing might look like.”
Australian singer-songwriter Didirri Peters, who records under Didirri, believes that artists and labels need to consider certifying and labelling music made by humans – a product he refers to as an “organic trademark”. The process, he says, could be similar to organic food certification or protected origin designation for regional cheeses and wines. “The farm certification process for organic food for instance: there are people who check, and it’s regulated.”
Peters suggests a certification process could include interviews with composers and producers about the origin of their music. “It’s pretty hard to answer the question of ‘how did you make this sound?’, if you didn’t make it.”
In the meantime, musicians still have live performance. Matt Walters from live music curation agency Sofar Sounds is best known for setting up Parlour Music, a platform that facilitates musicians to perform at house parties and in back yards. “I think these spaces in real life, that let us listen to a human being performing … will become more important,” he says.
Sofar Sounds explicitly curate the old fashioned way: by listening to the music. “It lives outside the algorithm,” Walters says. “I love intimate shows in these really interesting spaces.”
Peters agrees. “Live music is risky and people like the risk – they like going to see something unique. Not just unique artists, but unique as in: I was there, I saw that particular night.”